Is using splines in models where the response follows a distribution different from normal a good practice? Are there benefits when doing this?
@nipnipj totally! I work in biomedical field and applied researchers very often want to discretize their variables, when spline-type models would preserve all the available information. For variables like age for example, non-linear relationships are biologically plausible but would be completely missed using parametric models. I suppose the challenge is that for some discrete responses it might be difficult to have enough information to verify the reasonableness of the model, but that applies to parametric models too.